computational tools for designing and engineering enzymes - caver

Computational tools for designing and engineering enzymesJiri Damborsky and Jan Brezovsky

Available online at www.sciencedirect.com

ScienceDirect

Protein engineering strategies aimed at constructing enzymes

with novel or improved activities, specificities, and stabilities

greatly benefit from in silico methods. Computational methods

can be principally grouped into three main categories:

bioinformatics; molecular modelling; and de novo design.

Particularly de novo protein design is experiencing rapid

development, resulting in more robust and reliable predictions.

A recent trend in the field is to combine several computational

approaches in an interactive manner and to complement them

with structural analysis and directed evolution. A detailed

investigation of designed catalysts provides valuable

information on the structural basis of molecular recognition,

biochemical catalysis, and natural protein evolution.

Addresses

Loschmidt Laboratories, Department of Experimental Biology and

Research Centre for Toxic Compounds in the Environment, Faculty of

Science, Masaryk University, Kamenice 5/A13, 625 00 Brno, Czech

Republic

Corresponding author: Damborsky, Jiri ([email protected])

Current Opinion in Chemical Biology 2014, 19:8–16

This review comes from a themed issue on Biocatalysis and

biotransformation

Edited by Jeffrey C Moore and Uwe T Bornscheuer

1367-5931/$ – see front matter, # 2013 Elsevier Ltd. All rights

reserved.

http://dx.doi.org/10.1016/j.cbpa.2013.12.003

IntroductionEnzymes catalyse chemical reactions in living cells and

are used in a wide range of practical applications. In the

past, the applications had to be built around the limita-

tions of the enzyme; today, the enzymes can be

engineered to fit the process of interest [1��]. New

enzymes were traditionally obtained by isolating them

from native organisms. Later, enzymes were obtained

from metagenomic libraries without the need to culture

host organisms. DNA cloning technologies enabled

enzyme production in heterologous hosts and changes

to the genetic code to introduce modifications into the

protein structure. The invention of directed evolution

techniques opened new possibilities for massive or sys-

tematic mutagenesis.

More recently, focused directed evolution of selected

regions and the use of restricted genetic code have

become popular means of producing smaller and smarter


mutant libraries. Structural biology techniques such as

protein crystallography or NMR spectrometry allow the

determination of protein structures to atomic resolutions

and the employment of molecular modelling for identify-

ing mutagenesis hot spots. The most recent driving force

in the field stems from gene synthesis technology, which

allows the synthesis of gene coding for putative enzymes

from genetic databases, as well as the production of

computationally designed proteins.

In silico methods, ranging from bioinformatic analysis of

primary sequences, through computer simulations of ter-

tiary structures, to the prediction of novel structures by denovo design, wind through the platforms aimed at con-

structing optimal biocatalysts. We discuss the compu-

tational tools and their applications in protein design

that have been published in the past two years. We do

not cover the tools suitable for designing smart libraries

for focused directed evolution since we have reviewed

them recently elsewhere [2]. The article is structured in

three parts according to the purpose of the design: Firstly,

engineering enzyme activity; secondly, engineering

enzyme specificity; and finally, engineering enzyme

stability.

A number of excellent reviews have been published

recently on related topics. Davids and co-workers over-

viewed the methods suitable for designing focused

libraries and high-throughput screening [3]. Progress in

de novo protein design was discussed in the reviews by

Davey and Chica [4], Hilvert [5], Khare and Fleishman

[6], Kries and co-workers [7], and Wijma and Janssen [8].

Approaches for protein stabilization were covered by the

reviews of Wijma and co-workers [9], Bommarius and

Paye [10], Socha and Tokuriki [11], and Stepankova and

co-workers [12].

Designing and engineering enzyme activityBioinformatics

Suplatov and co-workers developed the web server

ZEBRA for analysing enzyme functional subfamilies

[13]. The server attempts to systematically identify and

analyse adaptive mutations. These subfamily specific

positions (SSPs) are conserved within the subfamily,

but should differ among them. The implemented stat-

istical analysis evaluates the significance of SSPs, which

can then be modified by rational design or focused

directed evolution. The method has been tested with

the a/b-hydrolase superfamily [14]. SSPs calculated for

the amidases were integrated into the sequence of the

lipase CALB and the library of mutants was constructed.

In silico screening of the library for the reactive

www.sciencedirect.com

[email protected]

http://dx.doi.org/10.1016/j.cbpa.2013.12.003

http://www.sciencedirect.com/science/journal/13675931

In silico tools for enzyme design Damborsky and Brezovsky 9

Figure 1

Molecular dynamics simulation of protein

Tunnel identification in individual structures

Tunnel clustering across all structures

Analysis of tunnel dynamics and importance

Identification of residues forming tunnel bottlenecks

Tu

nn

el w

idth

Simulation time

Current Opinion in Chemical Biology

Workflow for analysis of tunnels in dynamic protein structures using

CAVER [18�].

enzyme–substrate complexes resulted in the selection of

lipases with significantly improved amidase activity.

The JANUS method analyses multiple-sequence align-

ments to predict mutations required for interconversion

of structurally related but functionally distinct enzymes

[15]. The method has been validated by the intercon-

version of aspartate aminotransferase into tyrosine ami-

notransferase. The incorporation of 35 mutations

resulted in a protein with the desired specificity but

low catalytic activity, which had to be optimized by

DNA back-shuffling.

Yang and co-workers presented a computational approach

for engineering an allosteric regulation [16]. The authors

conducted a statistical comparison of catalytic and allo-

steric binding sites, which revealed that allosteric sites are

evolutionarily more variable and comprise more hydro-

phobic residues than the catalytic sites. The approach was

applied to the deregulation of the allostery in fructose-

1,6-bisphosphate, but it remains to be seen whether the

methodology will work for other enzyme families.

Molecular modelling

Biedermannova and co-workers combined several mol-

ecular modelling methods to study the effect of tunnel

mutations on kinetics and reaction mechanisms of

haloalkane dehalogenase [17]. The software tool CAVER[18�] was used to analyse tunnel dynamics in trajectories

obtained by molecular dynamic simulations (Figure 1)

and complemented with an analysis of products egressing

from buried active sites using Random Accelerated Molecu-lar Dynamics (RAMD). The energy barriers of the product

release, calculated by the Adaptive Biasing Force (ABF)

method, were in good agreement with the data from

transient kinetic experiments. A redesign of protein tun-

nels and gates [19] using dedicated software tools [20]

provides a useful strategy for engineering enzyme

activity.

De novo design

De novo protein design has become mainstream, with

more than half of the reviewed articles employing this

approach to some extent. This unprecedented research

activity makes de novo design more robust, accurate, and

reliable [6]. The software suites ROSETTA and ORBITare the most widely used, and web-based applications

were recently developed (see below). Designed enzymes

can catalyse non-biological reactions, including multistep

retroaldol transformation, Diels–Alder cycloaddition, and

proton transfer [5,7]. They typically do not meet the

efficiencies of natural enzymes, but can be improved

by directed evolution [21,22,23�,31��].

The methodology of computational protein design is

being continuously improved by the integration of novel

protocols. Hallen and co-workers introduced an algorithm

www.sciencedirect.com Current Opinion in Chemical Biology 2014, 19:8–16

10 Biocatalysis and biotransformation

called Dead-End Elimination with Perturbations (DEEPer)

for identifying global minimum-energy conformation of

structures with large backbone perturbations [24]. The

algorithm is expected to provide more realistic modelling

of backbone flexibility. Nivon and co-workers developed

the Pareto-Optimal Refinement Method for the efficient

design of initial scaffold libraries [25]. Keedy and co-

workers proposed a novel algorithm for modelling local

backrub motions, which are subtle backbone adjustments

taking place during amino acid substitutions [26]. These

motions participate in natural protein evolution and their

implementation in computational design algorithms

improves model accuracy. The incorporation of 114 non-

canonical amino acids into ROSETTA by building necess-

ary backbone-dependent rotamer libraries and the

parameterization and construction of a scoring function

were described by Renfrew and co-workers [27�].

A computational method to redesign the active site for

catalysing new reactions was developed by Khare and co-

workers [28�]. Using this method, the authors engineered

an organophosphate hydrolase starting from a functionally

diverse set of mononuclear zinc-containing metalloen-

zyme scaffolds. This redesigned enzyme showed the cat-

alytic efficiency kcat/Km of 104 M�1 s�1 after several rounds

of saturation and random mutagenesis, representing an

impressive increase in activity, greater than 107-fold.

Nosrati and Houk developed the software tool Selection ofActive/Binding Sites for Enzyme Redesign (SABER) to

analyse the functional sites of the proteins stored in

the Protein Data Bank [29]. The tool identifies the active

sites amendable to computational redesign by locating

potential catalytic residues in pre-defined spatial arrange-

ments. The software was thoroughly validated by its

identification of enzymes possessing the catalytic resi-

dues of o-succinyl benzoate synthase and designed Kemp

eliminase.

Privett and co-workers presented an iterative approach for

constructing a highly active enzyme catalysing Kemp

elimination reactions [30�]. The initial design was ana-

lysed by protein crystallography and molecular dynamics,

which revealed inactivity due to the presence of water

molecules in the active site and the high flexibility of the

active site residues. The mutagenesis focused deeper into

the interior of the protein and resulted in the enzyme

binding the transition state in an orientation flipped in

relation to the design model.

Blomberg and co-workers applied error-prone polymerase

chain reactions and DNA shuffling to identify hot spots

along the gene of designed Kemp eliminase, which were

subsequently mutagenized by focused mutagenesis

[31��,32]. The designed enzyme was evolved to acceler-

ate the chemical reaction 6 � 108-fold, approaching the

catalytic efficiency of highly optimized natural enzymes.


The crystal structure of the evolved enzyme was deter-

mined to a 1.09 A resolution and analysed using the

software tool CAVER [18�]. Future efforts to turn evolved

enzymes into perfect catalysts will focus on optimizing

protein dynamics.

Designing and engineering enzyme selectivityBioinformatics

The application of traditional bioinformatic tools for the

design of selective enzymes is difficult because primary

sequences do not contain sufficient information for

describing the spatial interactions of a substrate molecule

with the active site residues. Bioinformatic tools can be

used to identify interaction hot spots and restrict the

alphabet of substitutions for the design of smart libraries

in focused directed evolution [2], which is outside the

scope of this review.

Directed evolution experiments can be rationalized by

the development of predictive models trained on exper-

imental data, similar to Quantitative Structure-Activity

Relationships. Feng and co-workers developed the Adap-tive Substituent Reordering Algorithm (ASRA) which can be

employed in combination with many directed evolution

methods [33��]. ASRA identifies the underlying regularity

of the protein property landscape and makes predictions

about the properties of uncharacterized proteins

(Figure 2). Importantly, ASRA does not require assump-

tions of linearity, additivity, or any functional form of

structure–property relationships. The application of

ASRA was demonstrated with epoxide hydrolases, for

which the method provided reliable predictions of

multiple mutants with improved enantioselectivity. This

approach should also be applicable to other properties,

such as enzyme activity and stability.

Molecular modelling

Pratterr and co-workers applied molecular docking, mol-

ecular dynamics, and quantum mechanical calculations

for the rational redesign of the substrate ligand at the

metal centre of mononuclear non-heme iron(II)-depend-

ent hydrolase. Construction of only 10 enzyme variants

resulted in a novel enzyme showing a 9300-fold enantios-

electivity switch [34�]. Durmaz and co-workers employed

molecular docking, classical molecular dynamics, and

steered molecular dynamics for engineering the chain-

length specificity of lipase [35]. The single-point

mutation L360F lowered the activation barrier for

hydrolysis of C4 substrates, while the same mutation

increased this barrier for C8 substrates.

The empirical valence bond modelling technique was used

for a quantitative analysis of enantioselectivity of lipase

by Frushicheva and Warshel [36�]. The authors evaluated

both kcat and kcat/Km for individual enantiomers of 4-

nitrophenyl 2-methylheptanoate and concluded that

extensive sampling is essential for obtaining converging



Figure 2

Selection of positionsfor mutagenesis

Sampling of ruggedproperty landscape

100

75

50

25

0

100

75

50

25

0

100

75

50

25

0

W NM I

YW

WP

AC

TS

SQ

NM I

YW

SQ

KW

PA

CT

S

K

SP L

HE

A

AE

HL

PS

WPro

per

ty

Pro

per

ty

Pro

per

ty

Reordering of landscape toidentify a target region

Prediction of mutants with desiredproperty in the target region

Pos 1

Pos 1 Pos 1 Pos 1

Pos 2

Pos 2 Pos 2 Pos 2


Workflow for adaptive substituent reordering algorithm identifying mutants with desired properties [33��].

results. They further outlined the use of the linear responseapproximation approach to analyse the contributions of

each residue to the free energy corresponding to enzyme

enantioselectivity.

De novo design

The design of biocatalysts binding small ligands with

good affinity is a very challenging problem, requiring

precise calculation of rather weak protein–ligand inter-

actions. An optimized binding site must provide: Firstly,

specific energetically favourable hydrogen-bonding and

van der Waals interactions with the ligand; secondly, high

overall shape complementarity to the ligand; and finally,

structural pre-organization in the unbound protein state

to minimize entropy loss upon ligand binding. Tinberg

and co-workers [37��] demonstrated the first successful denovo design of ligand-binding protein with a low-to-mid

micromolar range (Figure 3). The designed binder of

Figure 3

Define specificinteractions

Place into suitable scaffoldand optimise binding site

Workflow for de novo design of ligand binding protein [37��].


digoxin could be further optimized by site-saturation

mutagenesis and selections using yeast surface display

and fluorescence-activated cell sorting, reaching a pico-

molar level of binding affinity and high selectivity.

A remarkable 116-fold improvement in the catalytic

activity of a-gliadin peptidase with gluten tetrapeptide

and an 877-fold switch in the enzyme’s substrate speci-

ficity were achieved by identifying an enzyme with pre-

existing activity and optimizing it with computational

protein design [38]. The engineered enzyme can be

produced with a good yield of 30 mg/L and can serve

as a potential oral therapeutic for celiac disease.

A new method for the computational design of binding

pockets for small molecules, POCKETOPTIMIZER, was

developed by Malisi and co-workers [39�]. The tool can

be used to modify the residues making up the protein

Filter out not complementarybinding sites

Select pre-organizedbinding sites

Obtain pre-organized bindingsite with high complementarity




binding pocket to improve or newly establish the binding

of a small ligand. The programme employs a protein–ligand scoring function for estimating the free energy of

binding using the molecular modelling programs AUTO-DOCK VINA and CADDSUITE, and AMBER to analyse

protein packing. The method was tested with a bench-

mark set consisting of proteins with available mutants

showing different binding affinities with their ligands and

known structures. The tool predicts correctly increased

affinity in 66% and 69% using CADDSUITE and AUTO-DOCK VINA, respectively.

Designing and engineering enzyme stabilityBioinformatics

A consensus-based design of thermostable proteins uses

evolutionary information from multiple sequence align-

ments to predict the most suitable — most often naturally

occurring — amino acids at a particular position. Anbar

and co-workers successfully applied a consensual

approach to engineering endoglucanase with a 14-fold

improved half-life at 85 8C [40]. The regions responsible

for improved thermostability could be identified by sub-

sequent molecular dynamic simulations. Blum and co-

workers combined structure-guided consensus with the B-FIT method to engineer thermostable a-amino ester

hydrolase with improved T50 by 7 8C and 1.3-fold

improved activity compared to wild-type enzyme [41].

Sullivan and co-workers addressed a problem with the low

reliability of the consensual approach by distinguishing

stabilizing from destabilizing mutations [42]. Consensus

mutations at more conserved positions were more likely

to be stabilizing in the model protein triosephosphate

isomerase, while mutations at highly correlated positions

were destabilizing. The authors suggest the exclusion of

the sites with high statistical correlations to other sites and

nearly invariant positions from the consensus design. The

application of this approach to the model protein

improved its thermostability by 8 8C. Wang and co-

workers developed the combinatorial coevolving-site satur-ation mutagenesis (CCSM) method for identifying hotspots

for mutagenesis [43]. The method targets functionally

correlated variation sites. It was validated with a-amylase,

for which thermal stability was improved by 8 8C. We

note that the underlying principle of CCSM is opposite to

the approach proposed by Sullivan and co-workers [42].

Construction of chimeric proteins is another well-estab-

lished approach of protein stabilization, mimicking the

process of DNA recombination. Romero and co-workers

demonstrated that the protein fitness landscape can be

efficiently inferred from experimental data using Gaussianprocesses [44��]. The fitness landscape describes how

protein contributes to organismal fitness, or it may

represent its biophysical properties, such as stability,

catalytic activity, and specificity. The authors developed

two different sequence design algorithms based on


Bayesian decision theory (Figure 4). The first algorithm

identifies small sets of sequences that are informative

about the landscape, and the second algorithm identifies

highly optimized sequences. Using these algorithms and

the data set of 261 sequences with known properties, the

authors engineered chimeric P450 enzymes that were

more thermostable than any mutants previously prepared

by chimeragenesis, rational design, or directed evolution.

The method should be applicable to modelling other

enzyme properties as well. A related approach was

employed to identify seven functional chimeras of argi-

nases from the SCHEMA-based library [45].

Shanmugaratnam and co-workers developed an interest-

ing approach allowing construction of chimeric proteins

combining the fragments from different protein folds

[46]. Using proteins belonging to flavodoxin-like and

(ba)8-barrel folds as a model system, the authors demon-

strated the power of recombination for diversifying

protein structures.

Molecular modelling

Tian and co-workers applied their software tools PRE-THERMUT to evaluate the free energy of unfolding and

PAREPRO to calculate the site evolutionary entropy of

methyl parathion hydrolase [47]. Seven selected positions

were saturated, leading to the identification of six pos-

itions capable of enhancing protein stability. A step-wise

recombination of these mutations resulted in a four-point

mutant with improved melting temperature (Tm) by

11.7 8C. A similar method, combining a consensual

approach with FOLDX calculation, provided cellobiohy-

drolase I with T50 improved by 4.7 8C [48]. Jacobs and co-

workers combined a consensus-based approach with POP-MUSIC calculations to design stable FN3 domains of two

different proteins [49]. The final constructs showed

increased stability, high expression levels, and good solu-

bility in Escherichia coli.

Raghunathan and co-workers proposed engineering sur-

face charges to modify protein stability and aggregation

properties [50]. The authors co-introduced stabilizing and

surface-charge modifying mutations and increased the

thermostability of a green fluorescent protein without

compromising its functional properties.

Kim and co-workers applied network analysis to the

protein structure for identifying hydrophobic interaction

clusters [51]. A network parameter of structural hierarchy,

k of k-clique, was used to predict stabilizing mutations of

xylanase. The melting temperature of the triple mutant

was increased by 8 8C, while retaining 74% of enzyme

activity. The ensemble-based Constraint Network Analysis(CNA) was used to investigate the relationship between

flexibility and thermostability of five citrate synthases by

Rathi and co-workers [52]. The authors showed a good

ability of CNA to identify a set of weak spots in protein



Figure 4

Collect experimentallycharacterized sequences

S1

S2

S4

S5

S3

S1

S1

d21

d31 d32

d42

d52

d43

d53 d54

d41

d51

S2

S2

S4

S4

S5

S5

S1

S2S3

S4

S5

S3

S3

Map sequences tohomologous structure

Characterized sequencesMost informative sequencesHighly optimized sequences uncertainty

Distance of sequences

Pro

per

ty

Establish residue-residuecontact map

Calculate sequence- andstructure-based distances

Create distancematrix of sequences


Workflow for property landscape analysis using the Gaussian process [44��].

structures that are suitable targets for stability pursuing

mutagenesis.

Koudelakova and co-workers demonstrated that as few as

four mutations introduced into the access tunnel of

haloalkane dehalogenase increased its thermostability

by 19 8C and extended the half-life in 40% dimethyl

sulfoxide from minutes to weeks [53�]. CAVER [18�]calculations were used to analyse protein tunnels and

molecular dynamic simulations were used to analyse

protein accessibility for the molecules of organic co-

solvents. FOLDX analysis carried for more than 220 000

mutations in 26 enzymes from all six enzyme classes

confirmed that the concept is generally applicable to

enzymes with buried active sites.

De novo design

Borgo and Havranek developed an automated protocol

ROSETTA VIP — void identification and packing — to

improve poorly packed protein cores [54]. The protocol

uses the ROSETTA HOLES analysis module and a simple

geometric scoring function to identify a small set of


mutations that may yield improved packing. The protocol

is applicable for stabilizing both designed and native

proteins against chemicals and thermal denaturation.

Protein WISDOM is a web-based tool integrating methods

for various protein design problems, including de novodesign [55]. The tool enables searching for templates,

designing optimized sequences with stability, analysing

fold specificity and binding affinity, and quantitative

assessment of the designs by ranking of sequences as

well as structures. EVODESIGN is another web-based tool

for designing optimal protein sequences of given scaffolds

while predicting multiple sequence and structure-based

features for design ranking [56�]. The tool uses MetropolisMonte-Carlo search of profiles constructed for homologous

structure families in the Protein Data Bank. The set of

local structure features, including secondary structures,

torsion angles, and solvation, are predicted by neural-

network training for optimization of structure packing.

These tools will make de novo protein design accessible to

a wider community due to their user-friendly web inter-

faces. Analogously, PYROSETTA Toolkit [57�] provides a



graphical user interface for preparing and running proto-

cols of ROSETTA and for data analysis.

Conclusions and outlookProtein engineering is one of the most dynamically devel-

oping scientific fields. The way proteins are being engin-

eered has changed dramatically over the last few decades,

primarily due to novel experimental technologies such

DNA cloning, high throughput and deep sequencing,

directed evolution methods, fluorescence-based sorting

technologies, and gene synthesis. In silico approaches

assisting protein design and engineering are being devel-

oped back to back with experimental techniques.

The bioinformatics approaches are most successfully used

for engineering protein stability. This is because low

resolution data and evolutionary information are suffi-

cient for identifying stabilizing mutations, but less suit-

able for predicting mutations determining enzyme

specificity or activity. Predictions of stable consensual,

ancestral, and chimeric sequences are particularly popu-

lar. The great challenge for bioinformatics analysis is the

continuous growth of sequences in genomic databases

and the urgent need for predicting a protein function from

a sequence alone. Analysis of sequences within enzyme

subfamilies and identification of SSPs, developed for

protein engineering applications, can possibly find a

use in these functional assignments. Newly developed

methods resembling quantitative-structure activity

relationships should allow the prediction of optimized

sequences from experimental data collected during

directed evolution experiments.

The molecular modelling approaches greatly benefit from

growing computational power and parallelized calcu-

lations on graphical cards. Larger molecular systems

can be studied by quantum mechanics and longer simu-

lation times can be achieved by molecular dynamics.

Molecular modelling studies often combine several insilico methods, including bioinformatics analysis, to

describe structure–function properties and predict

beneficial mutations. A typical example of this is the

prediction of thermostable proteins by combining the

calculation of Gibbs free energies with evolutionary

analyses. Current challenges include the quantitative

modelling of selectivities and activities, which require

the precise estimation of binding energies and reaction

activation barriers. The quality of force fields properly

describing the polarizability of atoms and conformational

behaviour of amino acid residues as well as extensive

sampling can be important for obtaining quantitative

results. We expect even closer integration of molecular

modelling methods with de novo design in near future.

The de novo protein design approach made the greatest

progress of the three reviewed categories over the last two

years. The approach was successfully used for the design


of enzymes catalysing single step and multi-step chemical

reactions, and for the design of a ligand binding site.

Graphical user interfaces and web servers are being

developed, making the de novo design accessible to a

wider community. The current challenges are sorting

successful designs from non-successful ones and design-

ing biocatalysts to match the catalytic efficiencies of

natural enzymes. De novo design should be also demon-

strated for a wider range of chemical reactions, including

those with more complex mechanisms. The design of

active sites for chemical reactions can be complemented

by the design of ligand transport pathways. Iterative

cycles of de novo design, structural analysis, and molecular

modelling can be further extended by transient kinetics

and studies of kinetic isotopic effects, dissecting individ-

ual reaction steps, and providing valuable mechanistic

information for further improvement of modelling proto-

cols. Designed enzymes need to be improved by many

rounds of directed evolution, and this will not change in

the near future.

AcknowledgementsThe authors are supported by grants from the European RegionalDevelopment Fund (CZ.1.05/2.1.00/01.0001), the Grant Agency of theCzech Republic (P503/12/0572), and the European Union FrameworkProgramme (NEWPROT 289350). The work of JB is supported by the‘Employment of the Best Young Scientists for International CooperationEmpowerment’ Programme (CZ1.07/2.3.00/30.0037), co-financed fromEuropean Social Fund and the state budget of the Czech Republic.

References and recommended readingPapers of particular interest, published within the period of review,have been highlighted as:

� of special interest�� of outstanding interest

1.��

Bornscheuer UT, Huisman GW, Kazlauskas RJ, Lutz S, Moore JC,Robins K: Engineering the third wave of biocatalysis. Nature2012, 485:185-194.

This review describes the major developments in protein engineering,including examples of successfully engineered proteins. New conceptsare highlighted and further challenges are outlined.

2. Sebestova E, Bendl J, Brezovsky J, Damborsky J: Computationaltools for designing smart libraries. In Directed Evolution LibraryCreation: Methods and Protocols. Edited by Gillam EMJ, Copp JN,Ackerley DF. New York: Humana Press; 2013.

3. Davids T, Schmidt M, Bottcher D, Bornscheuer UT: Strategies forthe discovery and engineering of enzymes for biocatalysis.Curr Opin Chem Biol 2013, 17:215-220.

4. Davey JA, Chica RA: Multistate approaches in computationalprotein design. Protein Sci 2012, 21:1241-1252.

5. Hilvert D: Design of protein catalysts. Annu Rev Biochem 2013,82:447-470.

6. Khare SD, Fleishman SJ: Emerging themes in thecomputational design of novel enzymes and protein–proteininterfaces. FEBS Lett 2013, 587:1147-1154.

7. Kries H, Blomberg R, Hilvert D: De novo enzymes bycomputational design. Curr Opin Chem Biol 2013, 17:221-228.

8. Wijma HJ, Janssen DB: Computational design gainsmomentum in enzyme catalysis engineering. FEBS J 2013,280:2948-2960.

9. Wijma HJ, Floor RJ, Janssen DB: Structure- and sequence-analysis inspired engineering of proteins for enhancedthermostability. Curr Opin Struct Biol 2013, 23:588-594.


http://refhub.elsevier.com/S1367-5931(13)00235-4/sbref0005


























10. Bommarius AS, Paye MF: Stabilizing biocatalysts. Chem SocRev 2013, 42:6534-6565.

11. Socha RD, Tokuriki N: Modulating protein stability – directedevolution strategies for improved protein function. FEBS J2013, 280:5582-5595.

12. Stepankova V, Bidmanova S, Koudelakova T, Prokop Z,Chaloupkova R, Damborsky J: Strategies for stabilization ofenzymes in organic solvents. ACS Catal 2013 http://dx.doi.org/10.1021/cs400684x.

13. Suplatov D, Kirilin E, Takhaveev V, Svedas V: Zebra: a web serverfor bioinformatic analysis of diverse protein families. J BiomolStruct Dyn 2013 http://dx.doi.org/10.1080/07391102.2013.834514.

14. Suplatov DA, Besenmatter W, Svedas VK, Svendsen A:Bioinformatic analysis of a/b-hydrolase fold enzymes revealssubfamily-specific positions responsible for discrimination ofamidase and lipase activities. Protein Eng Des Sel 2012, 25:689-697.

15. Addington TA, Mertz RW, Siegel JB, Thompson JM, Fisher AJ,Filkov V, Fleischman NM, Suen AA, Zhang C, Toney MD: Janus:prediction and ranking of mutations required for functionalinterconversion of enzymes. J Mol Biol 2013, 425:1378-1389.

16. Yang J-S, Seo SW, Jang S, Jung GY, Kim S: Rational engineeringof enzyme allosteric regulation through sequence evolutionanalysis. PLoS Comput Biol 2012, 8:e1002612.

17. Biedermannova L, Prokop Z, Gora A, Chovancova E, Kovacs M,Damborsky J, Wade RC: A single mutation in a tunnel to theactive site changes the mechanism and kinetics of productrelease in haloalkane dehalogenase LinB. J Biol Chem 2012,287:29062-29074.

18.�

Chovancova E, Pavelka A, Benes P, Strnad O, Brezovsky J,Kozlikova B, Gora A, Sustr V, Klvana M, Medek P et al.: CAVER3.0: a tool for the analysis of transport pathways in dynamicprotein structures. PLoS Comput Biol 2012, 8:e1002708.

The article introduced a new version of the software tool suitable foranalysis of tunnels and channels in dynamical protein structures. Thesoftware implements several new algorithms for tunnel identification andclustering.

19. Gora A, Brezovsky J, Damborsky J: Gates of enzymes. Chem Rev2013, 113:5871-5923.

20. Brezovsky J, Chovancova E, Gora A, Pavelka A, Biedermannova L,Damborsky J: Software tools for identification, visualizationand analysis of protein tunnels and channels. Biotechnol Adv2013, 31:38-49.

21. Khersonsky O, Kiss G, Roethlisberger D, Dym O, Albeck S,Houk KN, Baker D, Tawfik DS: Bridging the gaps in designmethodologies by evolutionary optimization of the stabilityand proficiency of designed Kemp eliminase KE59. Proc NatlAcad Sci U S A 2012, 109:10358-10363.

22. Althoff EA, Wang L, Jiang L, Giger L, Lassila JK, Wang Z, Smith M,Hari S, Kast P, Herschlag D et al.: Robust design andoptimization of retroaldol enzymes. Protein Sci 2012, 21:717-726.

23.�

Giger L, Caner S, Obexer R, Kast P, Baker D, Ban N, Hilvert D:Evolution of a designed retro-aldolase leads to completeactive site remodeling. Nat Chem Biol 2013, 9:494-498.

Random mutagenesis applied to the designed retro-aldolase resulted in4.4 � 103-fold activity improvement. Structural analysis of initial designand evolved variants revealed the importance of loop mobility andsupporting functional groups.

24. Hallen MA, Keedy DA, Donald BR: Dead-end elimination withperturbations (DEEPer): a provable protein design algorithmwith continuous sidechain and backbone flexibility. Proteins2013, 81:18-39.

25. Nivon LG, Moretti R, Baker D: A Pareto-optimal refinementmethod for protein design scaffolds. PLoS ONE 2013,8:e59004.

26. Keedy DA, Georgiev I, Triplett EB, Donald BR, Richardson DC,Richardson JS: The role of local backrub motions in evolvedand designed mutations. PLoS Comput Biol 2012, 8:e1002629.


27.�

Renfrew PD, Choi EJ, Bonneau R, Kuhlman B: Incorporation ofnoncanonical amino acids into Rosetta and use incomputational protein–peptide interface design. PLoS ONE2012, 7:e32637.

The article describes the incorporation of 114 noncanonical amino acidsinto the Rosetta modelling suite by building backbone-dependent rota-mer libraries and the parameterization and construction of a scoringfunction.

28.�

Khare SD, Kipnis Y, Greisen PJ, Takeuchi R, Ashani Y,Goldsmith M, Song Y, Gallaher JL, Silman I, Leader H et al.:Computational redesign of a mononuclear zincmetalloenzyme for organophosphate hydrolysis. Nat ChemBiol 2012, 8:294-300.

The authors converted a zinc-dependent adenosine deaminase into anorganophosphate hydrolase by de novo protein design and evolutionaryevolution. The very low activity of the scaffold protein was increased>107-fold.

29. Nosrati GR, Houk KN: SABER: a computational method foridentifying active sites for new reactions. Protein Sci 2012,21:697-706.

30.�

Privett HK, Kiss G, Lee TM, Blomberg R, Chica RA, Thomas LM,Hilvert D, Houk KN, Mayo SL: Iterative approach tocomputational enzyme design. Proc Natl Acad Sci U S A 2012,109:3790-3795.

The article describes the power of an iterative approach combining denovo design, molecular modelling, and protein crystallography. Thestructural analysis revealed binding of the transition state in the orienta-tion flipped to the design model.

31.��

Blomberg R, Kries H, Pinkas DM, Mittl PRE, Grutter MG,Privett HK, Mayo SL, Hilvert D: Precision is essential for efficientcatalysis in an evolved Kemp eliminase. Nature 2013 http://dx.doi.org/10.1038/nature12623.

This study demonstrates that effective placement of appropriate catalyticgroups in the right active site environment can bridge the efficiency gapbetween natural and designed enzymes. The evolved enzyme had kcat/Km

of 230 000 M�1 s�1 and kcat of 700 s�1, which is above the activity ofmany natural enzymes.

32. Hohne M, Bornscheuer UT: Protein engineering from ‘scratch’becoming mature. Angew Chem Int Ed Engl 2013 http://dx.doi.org/10.1002/anie.201309591.

33.��

Feng X, Sanchis J, Reetz MT, Rabitz H: Enhancing the efficiencyof directed evolution in focused enzyme libraries by theadaptive substituent reordering algorithm. Chem Eur J 2012,18:5646-5654.

The authors present a new method, complementing a directed evolution,which identifies the underlying regularity of the protein property land-scape and makes predictions about the property of uncharacterizedproteins.

34.�

Pratter SM, Konstantinovics C, Di Giuro CML, Leitner E, Kumar D,de Visser SP, Grogan G, Straganz GD: Inversion ofenantioselectivity of a mononuclear non-heme iron(II)-dependent hydroxylase by tuning the interplay of metal-centergeometry and protein structure. Angew Chem Int Ed Engl 2013,52:9677-9681.

Using molecular docking, molecular dynamics, and quantum mechanics,the authors redesigned the substrate ligand at the metal centre of a non-heme iron(II)-dependent hydrolase and achieved 9.3 � 103-fold change inenantioselectivity.

35. Durmaz E, Kuyucak S, Sezerman UO: Modifying the catalyticpreference of tributyrin in Bacillus thermocatenulatus lipasethrough in-silico modeling of enzyme–substrate complex.Protein Eng Des Sel 2013, 26:325-333.

36.�

Frushicheva MP, Warshel A: Towards quantitative computer-aided studies of enzymatic enantioselectivity: the case ofCandida antarctica lipase A. ChemBioChem 2012, 13:215-223.

The study describes the use of the empirical valence bond method andlinear response approximation for analysing the enantioselectivity oflipase. It is concluded that extensive sampling is essential for obtainingconverging results.

37.��

Tinberg CE, Khare SD, Dou J, Doyle L, Nelson JW, Schena A,Jankowski W, Kalodimos CG, Johnsson K, Stoddard BL et al.:Computational design of ligand-binding proteins with highaffinity and selectivity. Nature 2013, 501:212-216.

The authors demonstrate the de novo design of digoxenin-binding proteinwith micromolar affinity. The designed protein was further optimized by







http://dx.doi.org/10.1021/cs400684x

http://dx.doi.org/10.1021/cs400684x

http://dx.doi.org/10.1080/07391102.2013.834514

http://dx.doi.org/10.1080/07391102.2013.834514


































































http://dx.doi.org/10.1038/nature12623

http://dx.doi.org/10.1038/nature12623

http://dx.doi.org/10.1002/anie.201309591

http://dx.doi.org/10.1002/anie.201309591























saturation mutagenesis and yeast surface display, reaching a picomolarlevel of binding affinity.

38. Gordon SR, Stanley EJ, Wolf S, Toland A, Wu SJ, Hadidi D,Mills JH, Baker D, Pultz IS, Siegel JB: Computational design ofan alpha-gliadin peptidase. J Am Chem Soc 2012, 134:20513-20520.

39.�

Malisi C, Schumann M, Toussaint NC, Kageyama J, Kohlbacher O,Hocker B: Binding pocket optimization by computationalprotein design. PLoS ONE 2012, 7:e52505.

The article introduces a benchmark-set for binding pocket optimization.The set is used to develop a new method correctly predicting affinity-changing mutations with 69% reliability.

40. Anbar M, Gul O, Lamed R, Sezerman UO, Bayer EA: Improvedthermostability of Clostridium thermocellum endoglucanaseCel8A by using consensus-guided mutagenesis. Appl EnvironMicrobiol 2012, 78:3458-3464.

41. Blum JK, Ricketts MD, Bommarius AS: Improved thermostabilityof AEH by combining B-FIT analysis and structure-guidedconsensus method. J Biotechnol 2012, 160:214-221.

42. Sullivan BJ, Nguyen T, Durani V, Mathur D, Rojas S, Thomas M,Syu T, Magliery TJ: Stabilizing proteins from sequence statistics:the interplay of conservation and correlation in triosephosphateisomerase stability. J Mol Biol 2012, 420:384-399.

43. Wang C, Huang R, He B, Du Q: Improving the thermostability ofalpha-amylase by combinatorial coevolving-site saturationmutagenesis. BMC Bioinformatics 2012, 13:263.

44.��

Romero PA, Krause A, Arnold FH: Navigating the protein fitnesslandscape with Gaussian processes. Proc Natl Acad Sci U S A2013, 110:E193-E201.

The authors developed novel algorithms for modelling protein fitnesslandscapes using Gaussian processes, using them to predict chimericP450 with T50 of 69.7 8C, which is 8.7 8C more stable than the variantsobtained by directed evolution.

45. Romero PA, Stone E, Lamb C, Chantranupong L, Krause A,Miklos AE, Hughes RA, Fechtel B, Ellington AD, Arnold FH et al.:SCHEMA-designed variants of human arginase I and II revealsequence elements important to stability and catalysis. ACSSynth Biol 2012, 1:221-228.

46. Shanmugaratnam S, Eisenbeis S, Hocker B: A highly stableprotein chimera built from fragments of different folds. ProteinEng Des Sel 2012, 25:699-703.

47. Tian J, Wang P, Huang L, Chu X, Wu N, Fan Y: Improving thethermostability of methyl parathion hydrolase fromOchrobactrum sp. M231 using a computationally aidedmethod. Appl Microbiol Biotechnol 2013, 97:2997-3006.


48. Komor RS, Romero PA, Xie CB, Arnold FH: Highly thermostablefungal cellobiohydrolase I (Cel7A) engineered using predictivemethods. Protein Eng Des Sel 2012, 25:827-833.

49. Jacobs SA, Diem MD, Luo J, Teplyakov A, Obmolova G, Malia T,Gilliland GL, O’Neil KT: Design of novel FN3 domains with highstability by a consensus sequence approach. Protein Eng DesSel 2012, 25:107-117.

50. Raghunathan G, Sokalingam S, Soundrarajan N, Madan B,Munussami G, Lee S-G: Modulation of protein stability andaggregation properties by surface charge engineering. MolBiosyst 2013, 9:2379-2389.

51. Kim T, Joo JC, Yoo YJ: Hydrophobic interaction networkanalysis for thermostabilization of a mesophilic xylanase.J Biotechnol 2012, 161:49-59.

52. Rathi PC, Radestock S, Gohlke H: Thermostabilizing mutationspreferentially occur at structural weak spots with a highmutation ratio. J Biotechnol 2012, 159:135-144.

53.�

Koudelakova T, Chaloupkova R, Brezovsky J, Prokop Z,Sebestova E, Hesseler M, Khabiri M, Plevaka M, Kulik D, KutaSmatanova I et al.: Engineering enzyme stability and resistanceto an organic cosolvent by modification of residues in theaccess tunnel. Angew Chem Int Ed Engl 2013, 52:1959-1963.

The article introduces a new concept of protein stabilization by engineer-ing protein access tunnels. Four substitutions in the tunnel of dehalo-genase increased its thermostability by 19 8C and extended the half-life in40% DMSO from minutes to weeks.

54. Borgo B, Havranek JJ: Automated selection of stabilizingmutations in designed and natural proteins. Proc Natl Acad SciU S A 2012, 109:1494-1499.

55. Smadbeck J, Peterson MB, Khoury GA, Taylor MS, Floudas CA:Protein WISDOM: a workbench for in silico de novo design ofbiomolecules. J Vis Exp 2013 http://dx.doi.org/10.3791/50476.

56.�

Mitra P, Shultis D, Zhang Y: EvoDesign: de novo protein designbased on structural and evolutionary profiles. Nucleic AcidsRes 2013, 41:W273-W280.

The authors developed a web-based tool for designing optimal proteinsequences of given scaffolds with predictions of multiple sequences andstructure-based features for design ranking.

57.�

Adolf-Bryfogle J, Dunbrack RL Jr: The PyRosetta Toolkit: agraphical user interface for the Rosetta software suite. PLoSONE 2013, 8:e66856.

The article presents a graphical user interface for preparing and runningprotocols of Rosetta modelling suite. The toolkit makes de novo designaccessible to a wider scientific community.































































http://dx.doi.org/10.3791/50476







computational tools for designing and engineering enzymes - caver

Documents